Foundation model
Japanese 基盤モデル（ファウンデーションモデル）
A foundation model is a base AI model trained in advance on a very large amount of data, which many services and tasks are then built on. Its strengths and weaknesses carry over to everything built on top of it.
A foundation model is a large AI model trained first on a wide range of data, such as text and images. Companies do not rebuild it from scratch each time; they add features like chat or translation on top of it and turn it into their own service. It is like laying a building's foundation once and then putting several shops on top. Stanford researchers named BERT, DALL-E and GPT-3 as examples.
Watch out When many services share the same base, they also share its weaknesses. A wrong answer or bias in one model can show up across several services built on it. Also, when an AI service's terms say your input "isn't used to train foundation models", that is a promise about whether your data helps improve the base model, so check it for the plan you actually use.
Source: On the Opportunities and Risks of Foundation Models · What are Foundation Models? · Copilot in Microsoft 365 apps for home: your data and privacy · Enterprise data protection in Microsoft Copilot and Microsoft Copilot Chat (checked 2026-10-26)